barry@pico.math.ucla.edu (Barry Merriman) (11/16/90)
Further search for the truth: Here's some hotter numbers on my trivial Million Multiply benchmark from Sun (Thanks to Larry Wake): Sun 3/50 with MC68881 coprocessor (compiled with cc -O -f68881) 0.24 MFLOPS (Yes, its true, every 3/50 I've ever used has had a 68881 chip on it.) SPARCstation 1+ (not a current product, but similar to SLC) 3.2 MFLOPS And, here's a special guest appearance by...The SPARC 2! SPARCstation 2 ( using ``cc -O'', and SunOS 4.1.1) 10. MFLOPS (note: the timing line for this read 0.1u 0.0s 0:00 122% 0+124k 0+0io 0pf+0w---how does it run at 122% ?) Recall the other top stats: NeXTSTation: 3.3 MFLOPS RS/6000: 3.7 MFLOPS DECStation 3100: 4.0 MFLOPS ----------------------------- As others have pointed out, SPECmarks are probably superior to my benchmark. On the other hand, I didn't have time to type the SPECmark code in at the demo NeXT :-). And, I do consider my little code to be the purest measurement of ``1 MFLOP'', except for complications from the shell, the OS version, the compiler version, caches, etc---after all, if a machine can't do well on that, its not likely to do better on more complex floating point codes. -- Barry Merriman UCLA Dept. of Math UCLA Inst. for Fusion and Plasma Research barry@math.ucla.edu (Internet)
bostrov@storm.UUCP (Vareck Bostrom) (11/17/90)
While we're tossing up benchmark results, take a look at this: HP 9000/220 (68010 at 10 ?? MHz) 1010 dhrys/sec 0.001 MFLOPS Why not use a raytracer as a benchmark? They seem to have a good mix of floating point and integer (though they are a bit heaver on fp). Other weird benches: IBM PC/XT (8088 at 5 MHz) 491 dhrys/sec <0.001 MFLOPS IBM PC/jr (8088 at 5 MHz) 366 dhrys/sec <0.001 MFLOPS I haven't ever tested anything slower than a PC/jr.